DMGL-MDA: A dual-modal graph learning method for microbe-drug association prediction

Area of Science:

• Microbiology
• Pharmacology
• Computational Biology

Background:

• Microbe-drug interactions significantly influence human health.
• Predicting microbe-drug associations (MDAs) is essential for safe drug administration.
• Traditional experimental methods for MDA prediction are costly and time-consuming.

Purpose of the Study:

• To develop a novel computational approach for efficient and accurate prediction of microbe-drug associations (MDAs).
• To overcome limitations of existing graph neural network (GNN) models, such as over-smoothing and over-squashing, and issues with similarity matrix dependency.

Main Methods:

• Proposed a novel graph representation learning model named dual-modal graph learning for microbe-drug association prediction (DMGL-MDA).
• DMGL-MDA incorporates a dual-modal embedding module, a bipartite graph network embedding module, and a predictor module.
• Evaluated DMGL-MDA against state-of-the-art methods on two benchmark datasets using cross-validation.

Main Results:

• DMGL-MDA demonstrated superior performance compared to existing methods.
• Cross-validation confirmed the effectiveness of the proposed model.
• Ablation experiments and case studies further validated the model's predictive capabilities.

Conclusions:

• DMGL-MDA provides a robust and efficient computational solution for predicting microbe-drug associations.
• The model addresses key challenges in existing GNN-based approaches, offering improved accuracy and reliability.
• This work facilitates low-cost, high-throughput screening of potential MDAs, aiding drug development and personalized medicine.